Paper Number
AM3 

Session
AI and ML Based Rheological Characterization

Title
Modeling yielding and thixotropic response of complex fluids using physics informed neural networks

Presentation Date and Time
October 10, 2022 (Monday) 10:30

Track / Room
Track 6 / Mayfair

Authors (Click on name to view author profile)

  1. Rathinaraj, Joshua D. (Massachusetts Institute of Technology, Department of Mechanical Engineering)
  2. Lennon, Kyle R. (Massachusetts Institute of Technology, Chemical Engineering)
  3. Gonzalez Cadena, Miguel A. (Aramco Americas)
  4. Santra, Ashok (Aramco Americas)
  5. Swan, James W. (Massachusetts Institute of Technology, Chemical Engineering)
  6. McKinley, Gareth H. (Massachusetts Institute of Technology, Mechanical Engineering)

Author and Affiliation Lines (in printed abstract book)
Joshua D. Rathinaraj1, Kyle R. Lennon2, Miguel A. Gonzalez Cadena3, Ashok Santra3, James W. Swan2 and Gareth H. McKinley1
1Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02142; 2Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139; 3Aramco Americas, Houston, MA 77084

Speaker / Presenter 
Rathinaraj, Joshua D.

Keywords
AI based; gels; glasses; ML based; mud rheology; rheometry techniques

Text of Abstract
In general, aqueous fluids are preferred over non-aqueous fluids for oil and gas well drilling due to lower environmental footprints. Typically, a water-based drilling fluid formulation consist of clays for suspending barite and removing the cuttings generated during the drilling process. Rheological properties like plastic viscosity (PV), yield point (YP) and static gel strength are generally correlated to effective particulate suspension. Viscoelastic properties of such fluids also contribute to particulate suspension; therefore it is important to understand various rheological properties in detail. Clay based fluids form transient gels that exhibit complex yielding and thixotropic behavior, which makes it difficult to apply simple rheological constitutive models to describe their nonlinear viscoelastic properties. We measure the complex rheology of a canonical drilling mud based on laponite clay and study the thixotropic and yielding behavior by modeling the evolving microstructure using a Kelvin-Voigt–based Isotropic Kinematic Hardening (K-IKH) model. To facilitate a data-rich approach for describing the complex thixo-elasto-visco-plastic (TEVP) flow response of these clay dispersions, we replace the usual governing differential equation describing the evolution of the internal parameters in the K-IKH model (e.g., the scalar microstructural parameter and tensorial back strain) with a physics-informed neural network (PINN). The evolution of these internal parameters is often ascribed to simple governing differential equation in the underlying constitutive models and cannot be directly measured by experiments but are implicitly obtained by making modeling assumptions regarding the stress-deformation characteristics of the evolving microstructure. Replacing the conventional differential equations with PINN mitigates the need for typical simplifying assumptions such as linearity and provides more adaptability to describe the yielding and thixotropic behavior, especially at large stress or strain amplitudes.